fextract_t *fextract_pitch_energy_create(int frame_len, char *m_e_params) {
fextract_t *fex;
dsp_fextract_t *mfcc;
fex = (fextract_t *) rs_malloc(sizeof(fextract_t),"emotion feature extraction data");
fex->n_features = V2_N_FEATURES;
/* Abtastrate etc. ... */
fex->samplerate = SAMPLERATE;
fex->frame_len = frame_len; /* global frame length (in frames) */
fex->pitch = pitch_create(AC_GAUSS);
fex->frame_shift = fex->frame_len - ((fex->pitch->method == AC_GAUSS? 2 : 1 ) * fex->pitch->periodsPerWindow / fex->pitch->minimumPitch - fex->pitch->dt) * fex->samplerate ; /* global shift (in frames) */
fex->hnr = NULL;
fex->vq = NULL;
/* MFCCs ... */
mfcc = (dsp_fextract_t *) rs_malloc(sizeof(dsp_fextract_t), "feature extraction data");
mfcc->type = dsp_fextype_MFCC;
mfcc->version = DSP_MK_VERSION(1, 4);
if (!dsp_mfcc_create(mfcc, m_e_params)) {
rs_free(mfcc);
mfcc = NULL;
rs_warning("Could not initialize MFCC configuration!");
}
fex->mfcc=mfcc;
return(fex);
}
void EmoVoicePitch::transform_enter (ssi_stream_t &stream_in,
ssi_stream_t &stream_out,
ssi_size_t xtra_stream_in_num,
ssi_stream_t xtra_stream_in[]) {
if (!_cfg) {
_cfg = pitch_create (ssi_cast (pitch_method_t, _options.method));
_cfg->minimumPitch = ssi_cast (mx_real_t, _options.minfreq);
_cfg->maximumPitch = ssi_cast (mx_real_t, _options.maxfreq);
}
}
void EmoVoicePitch::transform (ITransformer::info info,
ssi_stream_t &stream_in,
ssi_stream_t &stream_out,
ssi_size_t xtra_stream_in_num,
ssi_stream_t xtra_stream_in[]) {
ssi_size_t sample_number = stream_in.num;
short *srcptr = ssi_pcast (short, stream_in.ptr);
mx_real_t *dstptr = ssi_pcast (mx_real_t, stream_out.ptr);
mx_real_t *p = pitch_calc (_cfg, srcptr, sample_number);
memcpy (dstptr, p, _cfg->nframes * sizeof (mx_real_t));
pitch_destroy (_cfg);
_cfg = pitch_create (ssi_cast (pitch_method_t, _options.method));
_cfg->minimumPitch = ssi_cast (mx_real_t, _options.minfreq);
_cfg->maximumPitch = ssi_cast (mx_real_t, _options.maxfreq);
free (p);
}
ssi_size_t EmoVoicePitch::getSampleNumberOut (ssi_size_t sample_number_in) {
// check if we already know the result
if (sample_number_in == _sample_number_in) {
return _sample_number_out;
}
if (!_cfg) {
_cfg = pitch_create (ssi_cast (pitch_method_t, _options.method));
_cfg->minimumPitch = ssi_cast (mx_real_t, _options.minfreq);
_cfg->maximumPitch = ssi_cast (mx_real_t, _options.maxfreq);
}
// calculate new sample number
pitch_method_t method = _cfg->method;
mx_real_t minimumPitch = _cfg->minimumPitch;
mx_real_t dt = _cfg->dt;
int periodsPerWindow = _cfg->periodsPerWindow;
if (method == AC_GAUSS)
periodsPerWindow *= 2;
mx_real_t x1 = (mx_real_t) 0.5/SAMPLERATE;
mx_real_t dt_window = periodsPerWindow / minimumPitch;
int nFrames;
mx_real_t t1;
int result = fitInFrame (method >= FCC_NORMAL ? 1 / minimumPitch + dt_window : dt_window, dt, &nFrames, &t1, x1, sample_number_in);
SSI_ASSERT (result);
if (method >= FCC_NORMAL)
nFrames = nFrames - 1;
_sample_number_out = nFrames;
return nFrames;
}
bool EffectSBSMS::Process()
{
if(!bInit) {
sbsms_init(4096);
bInit = TRUE;
}
bool bGoodResult = true;
//Iterate over each track
//Track::All is needed because this effect needs to introduce silence in the group tracks to keep sync
this->CopyInputTracks(Track::All); // Set up mOutputTracks.
TrackListIterator iter(mOutputTracks);
Track* t;
mCurTrackNum = 0;
double maxDuration = 0.0;
if(rateStart == rateEnd)
mTotalStretch = 1.0/rateStart;
else
mTotalStretch = 1.0/(rateEnd-rateStart)*log(rateEnd/rateStart);
// Must sync if selection length will change
bool mustSync = (mTotalStretch != 1.0);
t = iter.First();
while (t != NULL) {
if (t->GetKind() == Track::Label &&
(t->GetSelected() || (mustSync && t->IsSynchroSelected())) )
{
if (!ProcessLabelTrack(t)) {
bGoodResult = false;
break;
}
}
else if (t->GetKind() == Track::Wave && t->GetSelected() )
{
WaveTrack* leftTrack = (WaveTrack*)t;
//Get start and end times from track
mCurT0 = leftTrack->GetStartTime();
mCurT1 = leftTrack->GetEndTime();
//Set the current bounds to whichever left marker is
//greater and whichever right marker is less
mCurT0 = wxMax(mT0, mCurT0);
mCurT1 = wxMin(mT1, mCurT1);
// Process only if the right marker is to the right of the left marker
if (mCurT1 > mCurT0) {
sampleCount start;
sampleCount end;
start = leftTrack->TimeToLongSamples(mCurT0);
end = leftTrack->TimeToLongSamples(mCurT1);
WaveTrack* rightTrack = NULL;
if (leftTrack->GetLinked()) {
double t;
rightTrack = (WaveTrack*)(iter.Next());
//Adjust bounds by the right tracks markers
t = rightTrack->GetStartTime();
t = wxMax(mT0, t);
mCurT0 = wxMin(mCurT0, t);
t = rightTrack->GetEndTime();
t = wxMin(mT1, t);
mCurT1 = wxMax(mCurT1, t);
//Transform the marker timepoints to samples
start = leftTrack->TimeToLongSamples(mCurT0);
end = leftTrack->TimeToLongSamples(mCurT1);
mCurTrackNum++; // Increment for rightTrack, too.
}
sampleCount trackEnd = leftTrack->TimeToLongSamples(leftTrack->GetEndTime());
// SBSMS has a fixed sample rate - we just convert to its sample rate and then convert back
float srIn = leftTrack->GetRate();
float srSBSMS = 44100.0;
// the resampler needs a callback to supply its samples
resampleBuf rb;
sampleCount maxBlockSize = leftTrack->GetMaxBlockSize();
rb.block = maxBlockSize;
rb.buf = (audio*)calloc(rb.block,sizeof(audio));
rb.leftTrack = leftTrack;
rb.rightTrack = rightTrack?rightTrack:leftTrack;
rb.leftBuffer = (float*)calloc(maxBlockSize,sizeof(float));
rb.rightBuffer = (float*)calloc(maxBlockSize,sizeof(float));
rb.offset = start;
rb.end = trackEnd;
rb.ratio = srSBSMS/srIn;
rb.resampler = new Resampler(resampleCB, &rb);
// Samples in selection
sampleCount samplesIn = end-start;
// Samples for SBSMS to process after resampling
sampleCount samplesToProcess = (sampleCount) ((real)samplesIn*(srSBSMS/srIn));
// Samples in output after resampling back
sampleCount samplesToGenerate = (sampleCount) ((real)samplesToProcess * mTotalStretch);
sampleCount samplesOut = (sampleCount) ((real)samplesIn * mTotalStretch);
double duration = (mCurT1-mCurT0) * mTotalStretch;
if(duration > maxDuration)
maxDuration = duration;
TimeWarper *warper = NULL;
if (rateStart == rateEnd)
{
warper = new LinearTimeWarper(mCurT0, mCurT0,
mCurT1, mCurT0+maxDuration);
} else
{
warper = new LogarithmicTimeWarper(mCurT0, mCurT1,
rateStart, rateEnd);
}
SetTimeWarper(warper);
sbsmsInfo si;
si.rs = rb.resampler;
si.samplesToProcess = samplesToProcess;
si.samplesToGenerate = samplesToGenerate;
si.stretch0 = rateStart;
si.stretch1 = rateEnd;
si.ratio0 = pitchStart;
si.ratio1 = pitchEnd;
rb.sbsmser = sbsms_create(&samplesCB,&stretchCB,&ratioCB,rightTrack?2:1,quality,bPreAnalyze,true);
rb.pitch = pitch_create(rb.sbsmser,&si,srIn/srSBSMS);
rb.outputLeftTrack = mFactory->NewWaveTrack(leftTrack->GetSampleFormat(),
leftTrack->GetRate());
if(rightTrack)
rb.outputRightTrack = mFactory->NewWaveTrack(rightTrack->GetSampleFormat(),
rightTrack->GetRate());
sampleCount blockSize = SBSMS_FRAME_SIZE[quality];
rb.outBuf = (audio*)calloc(blockSize,sizeof(audio));
rb.outputLeftBuffer = (float*)calloc(blockSize*2,sizeof(float));
if(rightTrack)
rb.outputRightBuffer = (float*)calloc(blockSize*2,sizeof(float));
long pos = 0;
long outputCount = -1;
// pre analysis
real fracPre = 0.0f;
if(bPreAnalyze) {
fracPre = 0.05f;
resampleBuf rbPre;
rbPre.block = maxBlockSize;
rbPre.buf = (audio*)calloc(rb.block,sizeof(audio));
rbPre.leftTrack = leftTrack;
rbPre.rightTrack = rightTrack?rightTrack:leftTrack;
rbPre.leftBuffer = (float*)calloc(maxBlockSize,sizeof(float));
rbPre.rightBuffer = (float*)calloc(maxBlockSize,sizeof(float));
rbPre.offset = start;
rbPre.end = end;
rbPre.ratio = srSBSMS/srIn;
rbPre.resampler = new Resampler(resampleCB, &rbPre);
si.rs = rbPre.resampler;
long pos = 0;
long lastPos = 0;
long ret = 0;
while(lastPos<samplesToProcess) {
ret = sbsms_pre_analyze(&samplesCB,&si,rb.sbsmser);
lastPos = pos;
pos += ret;
real completion = (real)lastPos/(real)samplesToProcess;
if (TrackProgress(0,fracPre*completion))
return false;
}
sbsms_pre_analyze_complete(rb.sbsmser);
sbsms_reset(rb.sbsmser);
si.rs = rb.resampler;
}
// process
while(pos<samplesOut && outputCount) {
long frames;
if(pos+blockSize>samplesOut) {
frames = samplesOut - pos;
} else {
frames = blockSize;
}
outputCount = pitch_process(rb.outBuf, frames, rb.pitch);
for(int i = 0; i < outputCount; i++) {
rb.outputLeftBuffer[i] = rb.outBuf[i][0];
if(rightTrack)
rb.outputRightBuffer[i] = rb.outBuf[i][1];
}
pos += outputCount;
rb.outputLeftTrack->Append((samplePtr)rb.outputLeftBuffer, floatSample, outputCount);
if(rightTrack)
rb.outputRightTrack->Append((samplePtr)rb.outputRightBuffer, floatSample, outputCount);
double frac = (double)pos/(double)samplesOut;
int nWhichTrack = mCurTrackNum;
if(rightTrack) {
nWhichTrack = 2*(mCurTrackNum/2);
if (frac < 0.5)
frac *= 2.0; // Show twice as far for each track, because we're doing 2 at once.
else {
nWhichTrack++;
frac -= 0.5;
frac *= 2.0; // Show twice as far for each track, because we're doing 2 at once.
}
}
if (TrackProgress(nWhichTrack, fracPre + (1.0-fracPre)*frac))
return false;
}
rb.outputLeftTrack->Flush();
if(rightTrack)
rb.outputRightTrack->Flush();
leftTrack->ClearAndPaste(mCurT0, mCurT1, rb.outputLeftTrack,
true, false, GetTimeWarper());
if(rightTrack) {
rightTrack->ClearAndPaste(mCurT0, mCurT1, rb.outputRightTrack,
true, false, GetTimeWarper());
}
}
mCurTrackNum++;
}
else if (mustSync && t->IsSynchroSelected())
{
t->SyncAdjust(mCurT1, mCurT0 + (mCurT1 - mCurT0) * mTotalStretch);
}
//Iterate to the next track
t = iter.Next();
}
if (bGoodResult)
ReplaceProcessedTracks(bGoodResult);
// Update selection
mT0 = mCurT0;
mT1 = mCurT0 + maxDuration;
return bGoodResult;
}
pitch_t *hnr_create() {
pitch_t *hnr=pitch_create(AC_HANNING);
hnr=pitch_hnr_configure(hnr);
return hnr;
}
